Giving A.I. the Old College Try … Boosts Graduation Rates
WHY YOU SHOULD CARE
Graduation rates at universities in the U.S. have been plummeting for half a century. But artificial intelligence and data crunching are helping turn the tide.
Austin Birchel needed financial aid to afford Georgia State University. But even after winning a scholarship that covered 90 percent of his tuition, he was in for a shock when he checked his GSU account the summer before his freshman year. The scholarship wasn’t showing up. Within minutes, though, Birchell found that his Social Security number had been written incorrectly on one of the forms. The error was corrected, and his scholarship was applied. Thanks to artificial intelligence, he didn’t have to spend hours on the phone with the admissions office to figure out what went wrong.
Birchell is just one of thousands of university students at GSU and across the country who’ve benefited from AI and predictive analytics tools that are helping American higher education institutions fight — and slowly fix — what previously seemed intractable problems. College graduation rates in the United States have been plummeting for half a century. Today, nearly half of U.S. college students don’t complete their degree in six years. The graduation rates for minorities, low-income learners and first-generation students are even lower. Enrollment has also been declining for years. In 2018, there were 1.7 million fewer undergrad and graduate students in the U.S. than there were in 2011 — a decline of 9 percent, according to the National Student Clearinghouse.
By turning to predictive tools for help, at least five American universities are attempting to reach students and address their concerns faster, at times even before the students approach college authorities with their problems. And these tools are showing results.
We understand the power that AI has.
Bryce Kunkel, senior technology and marketing coordinator, University of Oklahoma
In 2015, GSU was struggling with its “summer melt” rate — students who are enrolled in the spring but fail to show up in the fall — which had gone up to 19 percent. The school’s chatbot, named Pounce after the school’s blue panther mascot, was launched in 2016 to target that crisis. It’s working. Pounce helps answer questions about financial aid, housing and registration. GSU has been able to reduce its summer melt by 37 percent in just two years.
In 2017, Southern Connecticut State University implemented an AI program using IBM’s Watson to create new student services based on data insights. The University of Texas uses a predictive analytics systems to rate each freshman’s likelihood of graduating so staff can offer support to those students who need it. Launched last year, Ivy, the tool used by the University of Oklahoma, logged more than 3,000 conversations between students and the bot in less than one year. Ivy uses natural language processing to answer current and prospective students’ questions online.
And at Southern Illinois University, the retention rate rose by 8.3 percent within a year of the college launching its predictive analytics tool, and has been “creeping up ever since,” says Provost Meera Komarraju. The tool, called EAB Navigate, identifies “early warning signs” that a student might be struggling. Are they absent from class? Turning in assignments late? Missing extracurriculars or sports? It then connects those students with a staff member who helps them get back on track. In some cases, staff work with university housing to physically track the student down and offer help. SIU’s mental health counseling services also reach out proactively to students.
“It’s not just academic issues, it’s social and emotional issues too,” Komarraju says. “Housing, academic affairs and tech are all working closely together.” SIU will launch another tool next month through AdmitHub that uses AI to offer more-advanced solutions to such problems.
For many years, a key reason for higher education’s recruitment and retention problem has been its failure to recognize that the student body itself is changing, experts say. “Nontraditional” students — those who are financially independent from their parents, have children, are employed full time or are attending school part-time — make up 74 percent of undergrads, according to North Carolina think tank RTI International. Many of these students have less time to navigate college bureaucracy or stick with a program they feel is unresponsive to their needs.
To fix its summer melt problem, GSU realized it had to find a different way to connect with students. “Email wasn’t working,” says Tim Renick, senior vice president for student success at GSU. “We would try to get them to complete the FAFSA [Free Application for Federal Student Aid], but it was too confusing.” School officials realized they were using the wrong platform to communicate with students: Nearly all students have smartphones and are texting, not emailing. GSU worked with Lindsay Page, an assistant professor of research methodology at the University of Pittsburgh, to develop Pounce. Simple answers are autogenerated; for more complex ones, a university staff member will reply. The bot hasn’t helped only in slashing summer melt numbers. At a time when reaching graduation day is taking longer and longer for most students, GSU was able to reduce the average graduation time by more than half a semester.
At the University of Oklahoma, administrators realized that accessing information — even on the university website — could be overwhelming for many current and potential students. By trawling the website, Ivy can categorize information and then, using natural language processing, give students the answers they’re seeking. “We understand the power that AI has,” says Bryce Kunkel, senior technology and marketing coordinator at the University of Oklahoma. “It’s not the same as just Googling, because we’re able to curate the answers more finely.” Kunkel says the number of calls to the admissions office has dropped significantly since Ivy was implemented.
Still, these technologies are far from perfect. “Right now, AI makes decisions based on what’s happened in the past,” says Jim Hendler, director of the Institute for Data Exploration and Applications at Rensselaer Polytechnic Institute in New York. “If it learns systemic bias, it will be enforced.” For instance, he says, if a company board has only White males, AI designed to shortlist the next board candidates would predict White males are the best fit. “Someone has to actually go in and train it specifically to look for diversity.” What if those biases make universities admit only students from backgrounds historically likelier to graduate because of institutional advantages they’ve enjoyed?
In the academic world, these tools are entering at a time the student body is more diverse than ever. That might be why so far these AI bots are further helping diversity. Page, who developed the GSU chatbot, says they’ve found that students from underserved backgrounds have benefited most. Asian American, Hispanic and African American students, as well as first-generation students or those on Pell Grants, have used Pounce more than the average GSU student has. Nearly 90 percent of GSU’s student body have some type of loan or grant, and the university is the largest enroller of adult learners in the country.
“Our students have complicated problems,” Renick says. “Is there a form that needs their father’s signature? Well, maybe they haven’t seen their father in six months.” Within two months of its launch, Pounce had answered 200,000 questions. Renick says several students report asking the chatbot questions they wouldn’t have asked a human because they were too embarrassed.
For these universities, understanding the changing face of the college student was the first step toward helping those students graduate. Using innovations such as AI is taking them even closer to that goal — as long as they can keep the technology bias-free.